Fail to construct preload hamt shards when traversal fails

[willscott]

Fix #54

[aschmahmann]

Don't feel strongly about the number chosen here, as mentioned in ipld/go-ipld-prime#531 we're really in a rough spot here.

This is a behavior change from previously, but if you don't want to return -1 or a different negative number then 0 is probably more reasonable than what we were doing here before.

[willscott]

i'm happy to take suggestions - do you think it makes sense to do -2 or similar in this case?

[rvagg]

I'd prefer 0 until we have some kind of agreement on the best way to deal with ipld/go-ipld-prime#531, otherwise this is an outlier.

[rvagg]

We could memoize the err inside UnixFSHAMTShard so the next call to a method that returns an error can return it. Unfortunately, by saying 0 to the Length() call we're minimising the chance that another method is going to be called on it.

[willscott]

given that 0 length leaves us unlikely to get the memoized error, i think i'm going to not add the complexity for that.